Increase in marine provinciality over the last 250 million years governed more by climate change than plate tectonics

Abstract

Amidst long-term fluctuations of the abiotic environment, the degree to which life organizes into distinct biogeographic provinces (provinciality) can reveal the fundamental drivers of global biodiversity. Our understanding of present-day biogeography implies that changes in the distribution of continents across climatic zones have predictable effects on habitat distribution, dispersal barriers and the evolution of provinciality. To assess marine provinciality through the Phanerozoic, here we (a) simulate provinces based on palaeogeographic reconstructions and global climate models and (b) contrast them with empirically derived provinces that we define using network analysis of fossil occurrences. Simulated and empirical patterns match reasonably well and consistently suggest a greater than 15% increase in provinciality since the Mesozoic era. Although both factors played a role, the simulations imply that the effect of the latitudinal temperature gradient has been twice as important in determining marine provinciality as continental configuration.

1. Introduction

Secular changes in global temperature [1] and the fragmentation of continents [2–5] had a large influence on marine biodiversity in deep time. However, the effects of both of these factors on the geographical distribution of life remain poorly documented, and they are complicated by incomplete and spatially heterogeneous fossil data [6]. Despite these limitations, there is evidence that marine biodiversity has increased over the course of the Phanerozoic, both at the global [6–9] and at the community level [10,11]. Provinciality, the partitioning of life into distinct biogeographic units, may also have increased through time [2]. This increase in provinciality has been invoked as the main driver of rising global diversity, especially in the Cenozoic era [2–4]. However, analyses of fossil data indicate that neither beta diversity [12] nor geodisparity, the association of compositional dissimilarity with geographical distance [13,14] have increased systematically over the course of the Phanerozoic.

Here we seek to answer the question, ‘Has provinciality increased through time, and, if so, why?’ We hypothesize a large-scale increase of provinciality since Pangean times (e.g. since the Triassic), because continents became more fragmented [4] providing opportunities for vicariance. Both higher connectivity of shelf area and a reduced latitudinal gradient of sea surface temperature (SST) were suggested to enhance biotic dispersal resulting in a more homogeneous global fauna [3]. Thus, an increase in continental fragmentation and an increase in the latitudinal SST gradient during climate cooling is expected to lead to greater provinciality, which would consequently result in greater global biodiversity [3,5] since the Early Mesozoic. The great unknown is how plate tectonic and climate change interact to create the high-level spatial organization of life.

Objectively delineating biogeographic units (‘provinces’ in this paper) is a challenging task due to the notoriously incomplete fossil record [6,7,15]. Traditional approaches that rely on index taxa [16], subjective endemism assessment [17], or time slice- and taxon-specific analyses [18–21] need to be replaced with multi-taxon approaches that offer a temporal perspective [22–24]. However, heterogeneous fossil coverage [6] can still distort biogeographic patterns (electronic supplementary material, figure S1.1). To circumvent sampling bias and quantify expected provinciality for the entire globe, we took advantage of the fact that modern marine biogeography [24,25] is almost fully explained by two variables, physical barriers and temperature gradients [3,24], which we use to reconstruct deep-time biogeography and provinciality.

Here we assess marine benthic provinciality over the Phanerozoic and demonstrate its increasing trajectory during the last 250 Myr with a simulation approach (figure 1a), which generates biogeographic partitionings directly from state-of-the-art palaeoclimate models [26] and palaeogeographic reconstructions [27]. To validate this result and document changes in marine biogeography, we also conducted an independent biogeographic partitioning analysis (figure 1b) based on fossil occurrence data from the Paleobiology Database (PaleoDB).

Figure 1.

Schemes for biogeographic partitioning. (a) Simulations using palaeoclimate models and palaeogeographic reconstructions. (b) Empirical approach using fossil occurrence data. (a) We defined climatic zones based on SST regimes in present-day provinces [24] by training a neural network (M) with present-day SSTs to predict the position of climatic zones from climate model outputs. We tabulated cells of the shelf habitat and calculated a distance matrix representing shortest marine connections. The shelf habitat was split into climate zones, in which a minimum spanning tree was built from the distance matrix of the corresponding cells. Province boundaries were inserted where the minimum distance was longer than a pre-set threshold. (b) The collections sampled in a time interval were assigned to equal-area geographical grid cells and species were tabulated in every spatio-temporal cell (heatmap indicates occurrence counts). The information was transformed to a graph, where every spatio-temporal cell was represented with a vertex. Vertices were connected if the spatio-temporal cells share species. The strength of connections was determined by the number of shared species and sampling [19]. The graph was partitioned with the ‘infomap’ community detection method, and the modules (i.e. provinces) were projected onto the original data structure. Bin numbers refer to time intervals and colours to fictional provinces (see electronic supplementary material, text S1 for details).

2. Methods

(a) . Simulations

The main drivers of modern biogeographic partitioning are the latitudinal SST gradient, as well as landmass or oceanic barriers to dispersal [24]. Accordingly, based on the principles of uniformitarianism, our model uses these determinants to generate simulated biogeographic partitionings. Palaeogeographic reconstructions over the Phanerozoic were taken from the PaleoMAP palaeocoastlines [28]. SST data were output from well-validated HadCM3 L climate models that perform comparably to fully state-of-the-art models [26] but are capable of performing the multi-millennial simulations required to spin-up the ocean.

We conducted simulations using a near-equal-area geographical grid [29] with a mean cell area of approximately 19 000 km². The approach is illustrated in figure 1a. Palaeoclimatic boundaries were determined by intersecting shallow sea areas with respective palaeotemperature reconstructions. This process is based on the observation that spatial patterns of global oceanic circulation are influenced by global mean temperature changes [30] and implicitly assumes that climatic zones and biogeographic provinces are defined by distinct temperature ranges. Recent marine biogeographic patterns [24] were transformed to indicate whether a province belongs to the tropical, temperate, or polar zone. We constructed a neural network (electronic supplementary material, text) to assess which zone a specific cell belonged to, based on present-day mean SSTs (HadCM3L output). Reconstructions of SST were then used to predict the distribution of climate zones in the past.

The procedure was applied separately to all palaeogeographic reconstructions, leading to 81 different biogeographic schemes over the Phanerozoic. Within a climate zone, we used minimum spanning trees of the water distances to describe potential paths between shelf habitat cells. Dispersal barriers were implemented using a threshold distance (d_t = 0.04 × the circumference of Earth, approx. 1600 km), if a minimum spanning tree connection was longer than this distance, the habitat space was assumed to be disjunct, and province boundaries were inserted. This choice of parameters was again informed by modern biogeographic patterns and assessed in sensitivity tests (electronic supplementary material, figure S1.7).

(b) . Biogeographic partitioning of fossil occurrence data

To delineate marine provinces using fossil occurrence data, we used well-preserved benthic groups including Brachiopoda, Bivalvia, Gastropoda, Bryozoa, Echinodermata, Anthozoa, Decapoda and Trilobita. Phanerozoic scale, species-level occurrence data from were downloaded from the Paleobiology Database (PBDB, http://www.paleobiodb.org) on 26 April, 2021. We assigned occurrences to geological stages using the timescale of Ogg et al. [31] in the same way as in the Phanerozoic scale analysis template of the R package ‘divDyn’ [32]. Large-scale fossil provinces were demarcated following Kocsis et al. [23,24] as described in the electronic supplementary material, text. Such network approaches have been previously described to be efficient in highlighting community changes across the Phanerozoic [33,34], and they were previously documented to be effective in defining modules that represent time-traceable biogeographic units (hereafter: provinces) for occurrences of recent benthic marine animals during the last 10 Myr [24].

(c) . Measuring provinciality

Provinciality is a single descriptor of geographical cells' membership in biogeographic provinces. We expressed provinciality using a metric that is numerically equivalent to Hurlbert's Probability of Interspecific Encounter (PIE) [35] and expresses the probability that two randomly selected cells belong to different biogeographic provinces:

$$\mathrm{PIE}=\frac{N}{N-1}\times \left(1-\sum _{i=1}^S{\left(\frac{N_i}{N}\right)}^2\right),$$ 2.1

where S is the number of detected provinces, N is the total number of geographical cells, and N_i is the number of cells that belong to province i.

(d) . Simulation post-processing

To make the simulated provinces comparable to empirical results, we resampled the simulation outputs to the same coarser-resolution grid that was used for fossil data analysis (electronic supplementary material, text S1). We omitted information from (i.e. masked) those cells of the simulation outputs that were not assigned to any province in the analysis of fossil data. Similarity of fossil data analysis and simulation results were assessed with the adjusted mutual information metric [36].

(e) . Determination of provinciality

Determinants of provinciality were assessed using both fossil data and simulation results (see electronic supplementary material, text for details). To contrast biogeographic patterns with observed patterns of diversity dynamics, we calculated sampling-standardized genus-level richness over the Phanerozoic (electronic supplementary material, figure S1.5) using the shareholder quorum subsampling algorithm [37] with the quorum of 0.6, as implemented in the ‘divDyn’ package [32]. We regressed observed provinciality values with total shelf area from the palaeogeographic reconstructions, mean SST of HadCM3L palaeoclimatic simulations and the continent fragmentation index of Zaffos et al. [4]. The variables were z-scored before the fitting of the statistical model. As the residuals of a simple linear model were significantly autocorrelated, the process was repeated with a generalized least squares (GLS) framework using an ARIMA model and separately with the first differences of the time series.

To better understand the determination of provinciality in the context of the simulation, two partial simulation sets were produced (electronic supplementary material, text), so that the contributions of SST change and of change of continent configuration to provinciality change could be measured independently. These two partial simulations produced provinciality values when either the latitudinal SST gradient or the continent configuration (palaeogeographic reconstruction) was held constant over time. The simulated provinciality was modelled using the provinciality series that was calculated from the partial simulation outputs, based on both the spatially complete and the fossil-sampled area.

3. Results

Our simulation based on the present-day continental configuration and the latitudinal SST gradient accurately replicates the spatial distribution of modern provinces (figure 2a, and electronic supplementary material, figures S1.3 and S2). Although we detect no significant increase over the whole Phanerozoic (rho = 0.14, 95% confidence interval: −0.09–0.37), provinciality increased by 16% in the last 250 Myr (figure 3). The strongest increase occurred from the Mid-Cretaceous to the Oligocene interval at a fourfold rate relative to this long-term trend.

Figure 2.

A selection of simulated (a) and observed (b) palaeobiogeographic maps in the last 250 Myr. The complete set of maps are available as electronic supplementary material (S2 for the simulation outputs, and S3 for data analysis). The colour scheme of panel (b) matches that of electronic supplementary material, figures S1.8 and S3. Colours on panel (a) are based on province sizes. Note that not all displayed observed clusters represent true provinces (see electronic supplementary material, text S1).

Figure 3.

Provinciality over the Phanerozoic. (a) Simulation results based on the complete globe (red) and based on areas that sampled in the fossil record (black). (b) Provinciality values observed from the fossil record. Dotted lines indicate linear regression models over the last 250 Myr. Geographical patterns of fossil sampling do not explain the low observed provinciality in the Triassic–Early Jurassic interval. (Online version in colour.)

As expected, the fossil-based (electronic supplementary material, figures S3) trajectory of marine benthic provinciality is more volatile than the simulated results, but it validates (figure 3b) the profound increase over the last 250 Myr (rho = 0.76, 0.62–0.87). This pattern is also supported by the number of observed provinces (modules), although this metric is less reliable (electronic supplementary material, text S1). Provinciality based on the fossil record is strongly correlated with global diversity over the last 250 Myr (rho = 0.64, 0.38–0.79). However, after accounting for temporal autocorrelation, we did not find evidence for a causal relationship.

Masking the simulation results to include only areas with fossil data (figure 3a) demonstrates the influence of spatially incomplete sampling on provinciality estimates. Although the simulated biogeographies only have a moderate match with fossil-derived schemes (median adjusted mutual information over the Phanerozoic = 0.39, figure 3), the resulting trajectories of provinciality are very similar (mean difference = 0.045), especially after the Early Jurassic (electronic supplementary material, figure S1.10). The simulations suggest greater-than-observed provinciality for the Triassic to Middle Jurassic interval (figure 3).

Varying the latitudinal SST gradient and continental configuration separately in simulations suggest that changes in the SST gradient had a more pronounced long-term effect on provinciality (Methods, electronic supplementary material, figure S1.6 and table S1.3). When the modern latitudinal SST gradient is kept constant over time, Mesozoic provincialities become even higher, and the post-Permian rise of provinciality disappears. This association underlines the pivotal role of temperature gradients in the evolution of marine biogeography. Simulations suggest that the SST gradient is at least twice (2.53 times) as important as continental configuration in controlling provinciality (model slopes: 1.85 versus 0.73; electronic supplementary material, table S1.3).

Fossil sampling patterns strongly influence which driving factors of provinciality can be observed (and electronic supplementary material, table S1.3) and on the accuracy of inferred biogeography: some ‘provinces’ clearly arise because of sampling noise (electronic supplementary material, text S1). The effects of climate change are nonetheless evident in these fossil-based biogeographic partitionings. Following the palaeolatitudinal distribution of shelf areas, Palaeozoic provinces are largely tropical, but Mesozoic and Cenozoic provinces are mostly subtropical (electronic supplementary material, figure S1.8). Provinces have a median duration of ca 25 Myr, and most present-day provinces can be traced back to the Palaeogene or even the K/Pg boundary. The faunas of the largest provinces (tropical Western Atlantic, European, tropical Indo-Pacific) emerged or spread after the Eocene–Oligocene boundary, coinciding with the onset of glaciation and the transition to icehouse climate conditions [38]. By the Early Miocene (ca 20 Ma), the modern biogeographic structure was fully established (figure 2).

4. Discussion

The basic pattern of increasing provinciality over the last 250 Myr is robust despite uncertainties regarding present-day processes and past patterns of biogeography. While the incomplete spatial coverage of fossil data [6] prevents the exact delineation of the size and number of provinces in the distant past, the observed increase of provinciality is confirmed by our simulation, which is robust with regard to free parameters and geographical resolution (electronic supplementary material, figures S1.2, S1.7). The mismatch between simulated and observed biogeographic provinciality partially stems from the fact that the simulations do not incorporate macroevolutionary phenomena. Mass extinctions, for example, may cause prolonged declines of provinciality [23], which could explain the depression of observed provinciality in the Triassic to Early Jurassic.

Our results support Valentine's [2,3] hypothesis on the outstanding role of geographical and climatic factors in shaping marine provincialism. The simulations suggest that lower than present-day values of provinciality in the Mesozoic can be partially attributed to a reduced latitudinal SST gradient, which is a consequence of higher global temperatures [39]. The extent of reduced provinciality in simulations may be underestimated because climate models still struggle to reproduce the full extent of the reduced SST gradients suggested from proxy data during greenhouse intervals [40]. Although thermal niches of species were documented to be conservative over millions of years [41], the discrepancy between simulated and observed provincialities might also attribute to differences in the responsiveness of Mesozoic and modern species distributions to temperature, which our uniformitarian simulation approach assumes.

The tropics already represent the largest climatic zone, so the link between their enlargement due to warming [30] and a consequent decrease in provinciality is not surprising. Larger dispersal distances are associated with low latitudes [42], which can enable these provinces to become larger when the tropics expand [30]. In addition, our results demonstrate that provinciality increases with cooling even without the increased number of latitudinal climatic zones that Valentine [3] originally suggested.

The effect of plate tectonics on biogeography is the foundation of vicariance biogeography and evident in phylogenetic patterns [43]. Our results confirm that continental fragmentation also has a direct effect on global provinciality. Two key events over the last 130 Myr were the complete opening of the Atlantic in the Mid-Cretaceous [44] and the onset of the Antarctic glaciation in the Oligocene [38]. Our simulations suggest that the former event led to a more profound increase in provinciality, whereas the latter led to modern marine provinces [24,25].

On Phanerozoic timescales, provinciality does not feature an overall increase, similar to beta diversity [12] and geodisparity [13,14]. However, when compared to provinciality, these variables exhibit different trajectories over the last 250 Myr. This difference might stem from the methods' different handling of high spatial turnover over short distances. The simulation approach does not consider within-province turnover at all, and the network analysis is effective in handling compositional overlap over faunal gradients [45], which make it well suited to unravel coarse-scale spatial patterns.

Increased provinciality remains a plausible mechanism for the global increase of marine biodiversity over the last 250 Myr [6,7]. Although the increase of provinciality over the same interval can be a contributing factor to the Cenozoic global diversification, a causal relationship remains difficult to quantify. Continent fragmentation was suggested to contribute to global diversity both by increasing provinciality and changes in sea-level [4,5]. Higher temperatures can also contribute to higher global diversity [1], despite the fact that long-term warming is associated with decreasing provinciality.

Despite the complexity of underlying processes, global provinciality can be simulated with just a few assumptions and limited environmental information. Our study demonstrates the strength of combining multiple types of data with simulations that model deep-time ecological patterns. These techniques can overcome limitations of data availability that arise from spatially incomplete record. Our results suggest that climate has an outstanding role in defining biogeographic patterns. Continental configuration changes slowly, but the fast pace of climatic changes can lead to a shift in the spatial arrangement of diversity [46] and perhaps a prolonged drop of provinciality.

Data accessibility

Electronic supplementary text, figures and tables, as well as code and data to replicate the analyses are deposited in Zenodo [47].

Authors' contributions

Á.T.K.: conceptualization, formal analysis, funding acquisition, investigation, methodology, project administration, software, validation, visualization, writing-original draft, writing-review and editing; C.J.R.: resources, software, writing-review and editing; C.R.S.: data curation, project administration, resources, writing-review and editing; P.V.: data curation, resources, writing-review and editing; W.K.: conceptualization, funding acquisition, project administration, resources, supervision, writing-review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Competing interests

The authors declare no competing interests.

Funding

The study was funded by the Deutsche Forschungsgemeinschaft (grant nos. Ko 5382/1-1, Ko 5382/1-2, Ko 5382/2-1, KI 806/16-1, AB 109/11-1) and is embedded in the Research Unit TERSANE (FOR 2332: Temperature-related stressors as a unifying principle in ancient extinctions). The palaeogeographic reconstructions were produced with support from the PALEOMAP Project Industrial Consortium (2003–2011).